Silicon Carbide Photonic Crystal Cavities with Integrated Color Centers
Greg Calusine1 , Alberto Politi1 , and David D. Awschalom1,2
1. Department of Physics, University of California, Santa Barbara, CA 93106, USA
2. Institute for Molecular Engineering, University of Chicago, Chicago, IL 60637, USA
Supplemental Materials
1. Photonic crystal band structure
The band structure of the 2D triangular lattice of holes that forms the photonic crystal was
studied using the MIT Photonic-Bands (MPB) package.S1 Figure S1 shows the first four
transverse electric(TE)-like modes of a silicon carbide(SiC) slab with hole radius r=0.29a and
thickness h=0.85a for a lattice constant a. The bandgap opens for frequencies between
approximately 0.3(2πc/a) and .38(2πc/a), where c is the speed of light. The resonant frequencies
of the simulated H1 and L3 cavities are highlighted by the corresponding dashed lines.
Fig. S1 Simulated band diagram (MPB) of the TE-like modes in a SiC 2D photonic crystal.
1
2. Fabrication process details
The starting material for photonic crystal fabrication consisted of a commercially
available 1 micron thick layer of <100> oriented 3C SiC grown on a 100 mm wafer of 500
micron thick <100> oriented p-type silicon(NovaSic). The film was unintentionally doped, with
nitrogen concentrations below 1x1016 per cubic centimeter and aluminum concentrations of less
than 1x1015 per cubic centimeter. This initial film is thinned down to a thickness of 300 nm, as
determined by standard ellipsometry techniques. Although the optimized H1 and L3 cavity
designs did not share the same simulated optimal structure thickness, this film thickness was
empirically determined to yield the highest measured cavity quality factors (Q) regardless of
exact cavity structure.
The surface of the material is initially polished to a roughness of .5 nm rms. Thinning of
the film using a 900 W power and 200 W bias SF6-based inductively coupled plasma(ICP) etch
increases the surface roughness to 1.2 nm rms. Fine tuning of the layer thickness can be achieved
by using a slower etch rate Ar/Cl etch process. This 500 W power and 200 W bias Ar/Cl (4:1,
respectively) ICP etch increases the surface roughness from .5nm rms to .8 nm rms. Samples
were surrounded with sacrificial pieces with no topside silicon surfaces exposed during thinning
etch steps because the physically aggressive 900 W power, 200 W bias SF6 etch was observed to
eject silicon nanoparticles from exposed silicon surfaces that could coat the sample and result in
micromasking. Additionally, these surrounding pieces maintained the homogeneity of the
plasma around the sample of interest in order to produce an etch that is uniform across the
entirety of the sample surface.
The ion implantation recipe described in the main text was chosen to produce a maximum
in the vacancy production versus depth at a position in the center of the thin film, as determined
using the Stopping Range of Ions in Matter (SRIM) program.S2 The recipe was found
empirically to produce a maximum in luminescence intensity from the Ky5 centers for a range of
implantation doses and annealing temperatures and times. Higher doses than 1x1013 ions per
square centimeter were observed to decrease the defect luminescence intensity, likely due to
residual lattice damage that remains even after the annealing step. The 7 degree implantation
angle was chosen to minimize ion channeling effects that SRIM does not account for in order to
produce a vacancy distribution that accurately matched the simulations.
The combination of process layers and etch steps used in the fabrication process was
chosen to optimize the structure sidewall angle and minimize degradation of the hard mask, and
for process consistency. The titanium capping layer minimized variation in the aluminum layer
composition by limiting oxidation. The AQUASave conductive polymer used in the electron
beam exposure step (JEOL 9300FS) minimized charging effects from uneven electrical contact
due to resist edge beads. The BCl3/Cl2 etch accurately transferred the resist pattern to the metal
hard mask without completely etching through the ZEP520 layer (selectivity of approximately
1:1). The SF6 etch immediately followed the BCl3/Cl2 etch without removing the sample from
2
the ICP chamber in order to avoid continued etching of the hard mask by unpassivated chlorine.
No change in the thickness of the hard mask layer was observed during the SF6 etch used to
transfer the pattern from the hard mask to the SiC. After the ICP etch, residues were observed on
the photonic crystal sidewalls. These residues disappeared after stripping the hard mask in
titanium and aluminum etchants and exposing the sample to buffered hydrofluoric acid. The
gaseous XeF2 isotropic etch undercuts the thin film by approximately 25 microns. Aside from
releasing the structure from the substrate, the undercut etch moves the silicon interface well
away from the confocal volume of the microscope, thereby eliminating background fluorescence
and laser scatter originating from the silicon substrate. This isotropic etch also eliminates any
silicon sputtered onto the hole sidewalls as a result of the SF6 etching into the underlying silicon
substrate.
3. Further experimental details
The structures were characterized in a home built scanning confocal microscope
equipped with a helium flow cryostat with optical access as depicted in figure S2. A flip mirror
in the excitation path allows for switching between a 1060-nm diode laser for off-resonant
excitation of the defects’ photoluminescence via the blue shifted absorption side-band and a
1085-to-1185-nm tunable 300-kHz linewidth Littmann-Metcalf diode laser for high resolution
cross-polarized resonant scattering spectroscopy. Laser excitation is passed through a 0.7
numerical aperture microscope objective lens mounted on a series of coarse scanning stages. A
fast-steering mirror (FSM) is incorporated into the optical path in order to alter the beam
incidence angle on the back aperture of the objective, thus allowing for fast spatial scanning with
a resolution of approximately 1 micron. Photoluminescence and scattered laser light are
collected through the same objective and reflected to a series of fiber couplers by a polarizing
beam splitter.
For photoluminescence measurements, the light emitted by the defects is long pass
filtered and coupled into a single-mode or multi-mode fiber. The light is then passed to either a
300 mm focal length spectrometer fitted with a liquid nitrogen cooled InGaAs CCD array for
taking broadband spectra with a resolution of ~ 1 nm or a 1 meter focal length spectrometer
fitted with the same CCD array with a resolution of ~ .1 nm. The light can also be collected out
of an alternate exit port on the 1 m focal length spectrometer and coupled to a superconducting
nanowire single photon detector through a single mode fiber. Stray excitation light that couples
into the collection path is filtered with a dichroic mirror and can be utilized for spatial mapping
of the photonic crystal cavity structure in order to place the laser excitation at the position of the
cavity (Figure 1(c)). All measurements were performed at a temperature of 20K.
For cross-polarized resonant scattering measurements, a vertically polarized laser beam is
reflected off a cavity rotated at 45 degrees with respect to the laser polarization axis. The cavity
3
scatters a fraction of the incident power into the far field with an orthogonal (horizontal)
polarization and it then passes into the collection path via a polarizing beam splitter. This light is
then collected through a single mode fiber that is coupled to a femtowatt photoreceiver. An
optical chopper wheel is used to modulate the incident laser intensity for lock-in detection.
Wavelength dependent power fluctuations are actively corrected for by a feedback loop
consisting of a power meter and automated variable neutral density (ND) filter. When the laser
wavelength is swept through the cavity resonance wavelength, a Fano lineshape is observed in
the scattered laser intensity. This method of cavity characterization complements
photoluminescence based methods in that it is capable of much higher resolution and can be used
to characterize cavities that lack incorporated emitters or are at room temperature. Due to silicon
carbide’s relatively small thermo-optic coefficientS3, only small ( ~ 1 nm) shifts in the cavity
resonance were observed between 20K and 295 K. Comparable structures in silicon typically
show about a factor of 10 times larger wavelength shift over this temperature range.S4
Fig. S2 Schematic picture showing the experimental setup used to characterize the SiC photonic
crystal devices.
4
4. Mode structure of H1 cavities
The H1 cavity mode exhibits another mode red shifted ~20 nm from the fundamental
mode as depicted in Figure S3. This mode typically exhibited a Q of ~ 1,000, similar to that of
the fundamental mode. Simulations of this mode yield a Q of approximately 3,000, a small
mode volume of .16 (𝜆/𝑛)3 and a mode profile depicted in Figure S4. While this mode does
exhibit a Q to mode volume ratio that is almost a factor of 10 higher than the fundamental for our
measured Q’s, the placement of the mode field maxima within the photonic crystal holes would
limit the achievable coupling of the mode to the optical transitions of color centers internal to the
film.
Fig. S3 Photoluminescence (PL) spectrum showing the fundamental H1 cavity mode at 1119 nm
and the red shifted mode at 1142 nm.
Fig. S4 Finite-difference time-domain (FDTD) simulation of red shifted mode’s Ex electric field
distribution.
5
5. Cavity imperfection simulations
Simulations of the effects of cavity imperfections on the structure Q’s were performed
with Lumerical’s FDTD Solutions software package. All simulation conditions were checked for
convergence and compared to results from the literature in order to verify their validity. To
include the effects of material absorption, the imaginary index of refraction of the thin film was
set to a value of 𝑛𝑖 = 7x10-5. This value was determined by relating the measured absorption
coefficient in reference 30 to the imaginary index of refraction through:
𝑛𝑖 =
𝛼𝜆
4𝜋
where 𝛼 is the absorption coefficient and 𝜆 is the cavity wavelength. The material used in
reference 30 originates from the same source as our material and is therefore likely to exhibit the
same degree of sub-band gap absorption.
To model fabrication imperfections, the initial, unperturbed structures were altered so
that their geometries reflected the various imperfections that can be introduced as a result of nonideal process steps. Sidewall angle was introduced by replacing the cylindrical photonic crystal
holes with conical holes with a lower radius corresponding to the unperturbed value and an upper
radius determined by the desired sidewall angle. In fabricated samples, this conical shape of the
photonic crystal holes results from a combination of the ICP etch process not being completely
vertical and degradation of the hard mask in the proximity of the holes during the etch. Figure S5
shows the dependence of the cavity Q’s on sidewall angle for all three cavity designs. As
discussed in the main text, at about 85 degrees sidewall angle the H1 cavity mode’s Q decreases
to below that of the optimized L3 design despite having a higher Q for the unperturbed structure.
This fact, in conjunction with the observed dependence of the structure Q’s on the other
imperfections, indicates that sidewall angle is likely the dominant factor limiting the Q’s of our
fabicated structures. Figure S5 also shows the dependence of the structure wavelength on
sidewall angle. The large shift in wavelength that results from nonvertical sidewalls can cause
the fundamental cavity mode to shift below the color centers’ zero phonon line. This shift needs
to be compensated for in order to place the mode within the defect emission band for coupling to
the internal emitters.
6
Fig. S5 Simulated dependences of cavity (a) Q factor and (b) resonant wavelength on sidewall
angle.
Random variations in the hole radii of the photonic crystal were introduced by adding or
subtracting small increments from the radii so that the values were normally distributed around a
mean corresponding to the structures’s ideal hole radius. The standard deviation of the Gaussian
distribution of radii was chosen to match the value of 1.5 nm measured with a scanning electron
microscope(SEM). This value corresponds to the resolution of the SEM and should be
considered an upper bound to the actual deviations. Random variations in the positions of the
photonic crystal holes were introduced in a similar manner. Again, the standard deviation of the
simulated distribution matched the measured distribution, which corresponded to the resolution
of the SEM. The Q values quoted in Table 1 are the mean of ten simulations for different random
configurations for each structure and deviation type. Table S1 compares the ideal cavity Q
values with the simulated mean Q values and standard errors.
Table SI. Simulated cavity Q’s for ideal structures and for structures with random radius and hole position variations.
Cavity Design
Ideal
1,660
Hole Radius
Variations
1,673
Radius Variation Q
Standard Error
10
Hole Position
Variations
1,634
Position Variation Q
Standard Error
15
L3
L3 optimized
17,084
16,383
403
16,090
326
H1 optimized
45,058
33,122
2,435
41,682
2249
S1
S2
S3
S4
S. G. Johnson and J. D. Joannopoulos, Optics Express 8, no. 3, 173-190 (2001)
J.F. Ziegler, Nucl. Instrum. Methods Phys. Res., Sect. B 2004, 219-220, 1027–1036.
S. Yamada, B.-S. Song, J. Upham, T. Asano, Y. Tanaka, and S. Noda, Opt. Express 20,
147890-14796 (2012).
N. Hauke, T. Zabel, K. Müller, M. Kaniber, A. Laucht, D. Bougeard,G. Abstreiter, J. J.
Finley, and Y. Arakawa, New Journal of Physics 12, 053005 (2010).
7